UNSUPERVISED MACHINE LEARNING FOR SEISMIC ANOMALY DETECTION: LOCAL OUTLIER FACTOR ALGORITHM TO INDONESIAN EARTHQUAKE DATA

Abstract

Indonesia's location on the "Ring of Fire" poses a high risk for seismic events. Addressing this, our study applied the Local Outlier Factor (LOF) algorithm for advanced seismic anomaly detection, crucial for geotectonic upheaval prediction. The LOF, adept at unsupervised learning in label-scarce datasets, analyzed data from the Indonesian Meteorology, Climatology, and Geophysical Agency, validated for integrity. Our approach, considering local density deviations, offered a refined alternative to conventional threshold-based detection, accommodating seismic data's intrinsic variability. The LOF algorithm successfully pinpointed anomalies, revealing unique seismic events unconstrained by geography or time. A comparative analysis underscored the LOF's superiority in recognizing local deviations and handling disparate data densities. These findings highlight the LOF's utility in strengthening seismic risk mitigation and anticipatory measures. The diverse anomalies identified, varying in magnitude and depth, reflect Indonesia's complex seismic interplay. To conclude, the LOF proves potent for anomaly detection, potentially elevating public safety and disaster preparedness. Future research will compare the LOF with other unsupervised methods, seeking to deepen seismic risk comprehension